Complete CLIP-seq workflow from raw FASTQ to ENCODE-compliant binding sites, single-nucleotide crosslink maps, annotated peaks, motifs, and (optionally) differential binding. Orchestrates the 12 decision-grade clip-seq skills with ENCODE-standard parameters: 10 nt UMIs for eCLIP / NNNXXXXNN for iCLIP, 3'-only adapter trimming (preserving R2 5' truncation = crosslink site -1), STAR end-to-end alignment with 0.04 mismatch ceiling (0.07 for PAR-CLIP T->C), UMI-unique deduplication, CLIPper or Skipper peak calling against SMInput, stringent log2 FC >= 3 AND -log10 p >= 3 thresholds, and IDR rescue + self-consistency both < 2.
# Preprocessing + alignment + QC
conda install -c bioconda umi_tools cutadapt fastp star samtools bowtie2 preseq picard multiqc rseqc bedtools
# Peak calling and downstream (CLIPper is NOT on bioconda -- the bioconda `clipper` is an
# unrelated R package. Install YeoLab CLIPper from source:
# git clone https://github.com/YeoLab/clipper && cd clipper && conda env create -f environment3.yml && pip install .)
conda install -c bioconda pureclip homer idr meme
# R / Bioconductor for annotation and differential
BiocManager::install(c('ChIPseeker','DEWSeq','TxDb.Hsapiens.UCSC.hg38.knownGene'))
Input data:
- Paired IP and size-matched input (SMInput) FASTQ, plus a STAR index and the genome FASTA
- A chromosome-sizes file and a BED of expressed 3'UTRs, for the GC-matched motif background. Motif discovery is SKIPPED without them: HOMER run without an explicit -fasta background first-order-scrambles the input and reports a spuriously U/AU-rich logo on any UV-CLIP library.
Tell your AI agent what you want to do: - "Run the full eCLIP pipeline from FASTQ to ENCODE-stringent peaks for my RBP and SMInput pair" - "Adapt for PAR-CLIP - raise STAR mismatch ceiling to 0.07 and use PARalyzer downstream" - "Run the iCLIP2 single-end variant; demultiplex NNNXXXXNN first" - "Repeat-binding RBP (MATR3) - use multi-mapper alignment and CLAM peak rescue" - "Add the differential binding step between treatment and control replicates with DEWSeq" - "End the pipeline at single-nucleotide crosslink sites for mCross motif registration" - "Run the QC five-gate diagnostic first; tell me which gate fails"
"Complete eCLIP run for RBFOX2: preprocess with 10 nt UMI + two-pass trim, align with STAR ENCODE block, dedupe with --method=unique, CLIPper + SMInput normalization, stringent log2 FC >= 3 / -log10 p >= 3, IDR across two replicates"
"Same but use Skipper for FASTKD2 - chrM peaks; CLIPper misses them"
"PAR-CLIP for HuR: raise STAR --outFilterMismatchNoverReadLmax to 0.07; use PARalyzer with Corcoran 2011 default parameters"
"iCLIP2 multiplexed library: demultiplex NNNXXXXNN library barcode first, then UMI extract the 5 random Ns"
"Repeat-binding MATR3: STAR --outFilterMultimapNmax 100 --outSAMmultNmax -1; CLAM EM rescue downstream"
"Just run the five QC gates on my existing dedup BAM - preprocessing retention, alignment rate, library complexity, FRiP, IDR"
"End at single-nt crosslink sites with PureCLIP; I'll feed them to mCross myself"
"After per-condition pipeline, run DEWSeq differential with type + condition + type:condition interaction"
"Switch from RBP target identification to m6A profiling - use clip-seq/m6a-clip (miCLIP2 + m6Aboost or GLORI)"
"Antibody unavailable - use clip-seq/stamp-antibody-free with STAMP and APOBEC1-only control"
-q 6 -m 18--method=unique~ type + condition + type:condition tests differential binding; ~ condition confounds with expression.